pgtable_32.c 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455
  1. /*
  2. * This file contains the routines setting up the linux page tables.
  3. * -- paulus
  4. *
  5. * Derived from arch/ppc/mm/init.c:
  6. * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
  7. *
  8. * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
  9. * and Cort Dougan (PReP) (cort@cs.nmt.edu)
  10. * Copyright (C) 1996 Paul Mackerras
  11. *
  12. * Derived from "arch/i386/mm/init.c"
  13. * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
  14. *
  15. * This program is free software; you can redistribute it and/or
  16. * modify it under the terms of the GNU General Public License
  17. * as published by the Free Software Foundation; either version
  18. * 2 of the License, or (at your option) any later version.
  19. *
  20. */
  21. #include <linux/kernel.h>
  22. #include <linux/module.h>
  23. #include <linux/types.h>
  24. #include <linux/mm.h>
  25. #include <linux/vmalloc.h>
  26. #include <linux/init.h>
  27. #include <linux/highmem.h>
  28. #include <linux/memblock.h>
  29. #include <linux/slab.h>
  30. #include <asm/pgtable.h>
  31. #include <asm/pgalloc.h>
  32. #include <asm/fixmap.h>
  33. #include <asm/io.h>
  34. #include "mmu_decl.h"
  35. unsigned long ioremap_base;
  36. unsigned long ioremap_bot;
  37. EXPORT_SYMBOL(ioremap_bot); /* aka VMALLOC_END */
  38. #if defined(CONFIG_6xx) || defined(CONFIG_POWER3)
  39. #define HAVE_BATS 1
  40. #endif
  41. #if defined(CONFIG_FSL_BOOKE)
  42. #define HAVE_TLBCAM 1
  43. #endif
  44. extern char etext[], _stext[];
  45. #ifdef HAVE_BATS
  46. extern phys_addr_t v_mapped_by_bats(unsigned long va);
  47. extern unsigned long p_mapped_by_bats(phys_addr_t pa);
  48. void setbat(int index, unsigned long virt, phys_addr_t phys,
  49. unsigned int size, int flags);
  50. #else /* !HAVE_BATS */
  51. #define v_mapped_by_bats(x) (0UL)
  52. #define p_mapped_by_bats(x) (0UL)
  53. #endif /* HAVE_BATS */
  54. #ifdef HAVE_TLBCAM
  55. extern unsigned int tlbcam_index;
  56. extern phys_addr_t v_mapped_by_tlbcam(unsigned long va);
  57. extern unsigned long p_mapped_by_tlbcam(phys_addr_t pa);
  58. #else /* !HAVE_TLBCAM */
  59. #define v_mapped_by_tlbcam(x) (0UL)
  60. #define p_mapped_by_tlbcam(x) (0UL)
  61. #endif /* HAVE_TLBCAM */
  62. #define PGDIR_ORDER (32 + PGD_T_LOG2 - PGDIR_SHIFT)
  63. pgd_t *pgd_alloc(struct mm_struct *mm)
  64. {
  65. pgd_t *ret;
  66. /* pgdir take page or two with 4K pages and a page fraction otherwise */
  67. #ifndef CONFIG_PPC_4K_PAGES
  68. ret = kzalloc(1 << PGDIR_ORDER, GFP_KERNEL);
  69. #else
  70. ret = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
  71. PGDIR_ORDER - PAGE_SHIFT);
  72. #endif
  73. return ret;
  74. }
  75. void pgd_free(struct mm_struct *mm, pgd_t *pgd)
  76. {
  77. #ifndef CONFIG_PPC_4K_PAGES
  78. kfree((void *)pgd);
  79. #else
  80. free_pages((unsigned long)pgd, PGDIR_ORDER - PAGE_SHIFT);
  81. #endif
  82. }
  83. __init_refok pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
  84. {
  85. pte_t *pte;
  86. extern int mem_init_done;
  87. extern void *early_get_page(void);
  88. if (mem_init_done) {
  89. pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
  90. } else {
  91. pte = (pte_t *)early_get_page();
  92. if (pte)
  93. clear_page(pte);
  94. }
  95. return pte;
  96. }
  97. pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
  98. {
  99. struct page *ptepage;
  100. gfp_t flags = GFP_KERNEL | __GFP_REPEAT | __GFP_ZERO;
  101. ptepage = alloc_pages(flags, 0);
  102. if (!ptepage)
  103. return NULL;
  104. pgtable_page_ctor(ptepage);
  105. return ptepage;
  106. }
  107. void __iomem *
  108. ioremap(phys_addr_t addr, unsigned long size)
  109. {
  110. return __ioremap_caller(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED,
  111. __builtin_return_address(0));
  112. }
  113. EXPORT_SYMBOL(ioremap);
  114. void __iomem *
  115. ioremap_wc(phys_addr_t addr, unsigned long size)
  116. {
  117. return __ioremap_caller(addr, size, _PAGE_NO_CACHE,
  118. __builtin_return_address(0));
  119. }
  120. EXPORT_SYMBOL(ioremap_wc);
  121. void __iomem *
  122. ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags)
  123. {
  124. /* writeable implies dirty for kernel addresses */
  125. if (flags & _PAGE_RW)
  126. flags |= _PAGE_DIRTY | _PAGE_HWWRITE;
  127. /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
  128. flags &= ~(_PAGE_USER | _PAGE_EXEC);
  129. #ifdef _PAGE_BAP_SR
  130. /* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format
  131. * which means that we just cleared supervisor access... oops ;-) This
  132. * restores it
  133. */
  134. flags |= _PAGE_BAP_SR;
  135. #endif
  136. return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
  137. }
  138. EXPORT_SYMBOL(ioremap_prot);
  139. void __iomem *
  140. __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
  141. {
  142. return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
  143. }
  144. void __iomem *
  145. __ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags,
  146. void *caller)
  147. {
  148. unsigned long v, i;
  149. phys_addr_t p;
  150. int err;
  151. /* Make sure we have the base flags */
  152. if ((flags & _PAGE_PRESENT) == 0)
  153. flags |= PAGE_KERNEL;
  154. /* Non-cacheable page cannot be coherent */
  155. if (flags & _PAGE_NO_CACHE)
  156. flags &= ~_PAGE_COHERENT;
  157. /*
  158. * Choose an address to map it to.
  159. * Once the vmalloc system is running, we use it.
  160. * Before then, we use space going down from ioremap_base
  161. * (ioremap_bot records where we're up to).
  162. */
  163. p = addr & PAGE_MASK;
  164. size = PAGE_ALIGN(addr + size) - p;
  165. /*
  166. * If the address lies within the first 16 MB, assume it's in ISA
  167. * memory space
  168. */
  169. if (p < 16*1024*1024)
  170. p += _ISA_MEM_BASE;
  171. #ifndef CONFIG_CRASH_DUMP
  172. /*
  173. * Don't allow anybody to remap normal RAM that we're using.
  174. * mem_init() sets high_memory so only do the check after that.
  175. */
  176. if (mem_init_done && (p < virt_to_phys(high_memory)) &&
  177. !(__allow_ioremap_reserved && memblock_is_region_reserved(p, size))) {
  178. printk("__ioremap(): phys addr 0x%llx is RAM lr %p\n",
  179. (unsigned long long)p, __builtin_return_address(0));
  180. return NULL;
  181. }
  182. #endif
  183. if (size == 0)
  184. return NULL;
  185. /*
  186. * Is it already mapped? Perhaps overlapped by a previous
  187. * BAT mapping. If the whole area is mapped then we're done,
  188. * otherwise remap it since we want to keep the virt addrs for
  189. * each request contiguous.
  190. *
  191. * We make the assumption here that if the bottom and top
  192. * of the range we want are mapped then it's mapped to the
  193. * same virt address (and this is contiguous).
  194. * -- Cort
  195. */
  196. if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ )
  197. goto out;
  198. if ((v = p_mapped_by_tlbcam(p)))
  199. goto out;
  200. if (mem_init_done) {
  201. struct vm_struct *area;
  202. area = get_vm_area_caller(size, VM_IOREMAP, caller);
  203. if (area == 0)
  204. return NULL;
  205. area->phys_addr = p;
  206. v = (unsigned long) area->addr;
  207. } else {
  208. v = (ioremap_bot -= size);
  209. }
  210. /*
  211. * Should check if it is a candidate for a BAT mapping
  212. */
  213. err = 0;
  214. for (i = 0; i < size && err == 0; i += PAGE_SIZE)
  215. err = map_page(v+i, p+i, flags);
  216. if (err) {
  217. if (mem_init_done)
  218. vunmap((void *)v);
  219. return NULL;
  220. }
  221. out:
  222. return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
  223. }
  224. EXPORT_SYMBOL(__ioremap);
  225. void iounmap(volatile void __iomem *addr)
  226. {
  227. /*
  228. * If mapped by BATs then there is nothing to do.
  229. * Calling vfree() generates a benign warning.
  230. */
  231. if (v_mapped_by_bats((unsigned long)addr)) return;
  232. if (addr > high_memory && (unsigned long) addr < ioremap_bot)
  233. vunmap((void *) (PAGE_MASK & (unsigned long)addr));
  234. }
  235. EXPORT_SYMBOL(iounmap);
  236. int map_page(unsigned long va, phys_addr_t pa, int flags)
  237. {
  238. pmd_t *pd;
  239. pte_t *pg;
  240. int err = -ENOMEM;
  241. /* Use upper 10 bits of VA to index the first level map */
  242. pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
  243. /* Use middle 10 bits of VA to index the second-level map */
  244. pg = pte_alloc_kernel(pd, va);
  245. if (pg != 0) {
  246. err = 0;
  247. /* The PTE should never be already set nor present in the
  248. * hash table
  249. */
  250. BUG_ON((pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE)) &&
  251. flags);
  252. set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT,
  253. __pgprot(flags)));
  254. }
  255. return err;
  256. }
  257. /*
  258. * Map in a chunk of physical memory starting at start.
  259. */
  260. void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
  261. {
  262. unsigned long v, s, f;
  263. phys_addr_t p;
  264. int ktext;
  265. s = offset;
  266. v = PAGE_OFFSET + s;
  267. p = memstart_addr + s;
  268. for (; s < top; s += PAGE_SIZE) {
  269. ktext = ((char *) v >= _stext && (char *) v < etext);
  270. f = ktext ? PAGE_KERNEL_TEXT : PAGE_KERNEL;
  271. map_page(v, p, f);
  272. #ifdef CONFIG_PPC_STD_MMU_32
  273. if (ktext)
  274. hash_preload(&init_mm, v, 0, 0x300);
  275. #endif
  276. v += PAGE_SIZE;
  277. p += PAGE_SIZE;
  278. }
  279. }
  280. void __init mapin_ram(void)
  281. {
  282. unsigned long s, top;
  283. #ifndef CONFIG_WII
  284. top = total_lowmem;
  285. s = mmu_mapin_ram(top);
  286. __mapin_ram_chunk(s, top);
  287. #else
  288. if (!wii_hole_size) {
  289. s = mmu_mapin_ram(total_lowmem);
  290. __mapin_ram_chunk(s, total_lowmem);
  291. } else {
  292. top = wii_hole_start;
  293. s = mmu_mapin_ram(top);
  294. __mapin_ram_chunk(s, top);
  295. top = memblock_end_of_DRAM();
  296. s = wii_mmu_mapin_mem2(top);
  297. __mapin_ram_chunk(s, top);
  298. }
  299. #endif
  300. }
  301. /* Scan the real Linux page tables and return a PTE pointer for
  302. * a virtual address in a context.
  303. * Returns true (1) if PTE was found, zero otherwise. The pointer to
  304. * the PTE pointer is unmodified if PTE is not found.
  305. */
  306. int
  307. get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
  308. {
  309. pgd_t *pgd;
  310. pud_t *pud;
  311. pmd_t *pmd;
  312. pte_t *pte;
  313. int retval = 0;
  314. pgd = pgd_offset(mm, addr & PAGE_MASK);
  315. if (pgd) {
  316. pud = pud_offset(pgd, addr & PAGE_MASK);
  317. if (pud && pud_present(*pud)) {
  318. pmd = pmd_offset(pud, addr & PAGE_MASK);
  319. if (pmd_present(*pmd)) {
  320. pte = pte_offset_map(pmd, addr & PAGE_MASK);
  321. if (pte) {
  322. retval = 1;
  323. *ptep = pte;
  324. if (pmdp)
  325. *pmdp = pmd;
  326. /* XXX caller needs to do pte_unmap, yuck */
  327. }
  328. }
  329. }
  330. }
  331. return(retval);
  332. }
  333. #ifdef CONFIG_DEBUG_PAGEALLOC
  334. static int __change_page_attr(struct page *page, pgprot_t prot)
  335. {
  336. pte_t *kpte;
  337. pmd_t *kpmd;
  338. unsigned long address;
  339. BUG_ON(PageHighMem(page));
  340. address = (unsigned long)page_address(page);
  341. if (v_mapped_by_bats(address) || v_mapped_by_tlbcam(address))
  342. return 0;
  343. if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
  344. return -EINVAL;
  345. __set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0);
  346. wmb();
  347. flush_tlb_page(NULL, address);
  348. pte_unmap(kpte);
  349. return 0;
  350. }
  351. /*
  352. * Change the page attributes of an page in the linear mapping.
  353. *
  354. * THIS CONFLICTS WITH BAT MAPPINGS, DEBUG USE ONLY
  355. */
  356. static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
  357. {
  358. int i, err = 0;
  359. unsigned long flags;
  360. local_irq_save(flags);
  361. for (i = 0; i < numpages; i++, page++) {
  362. err = __change_page_attr(page, prot);
  363. if (err)
  364. break;
  365. }
  366. local_irq_restore(flags);
  367. return err;
  368. }
  369. void kernel_map_pages(struct page *page, int numpages, int enable)
  370. {
  371. if (PageHighMem(page))
  372. return;
  373. change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
  374. }
  375. #endif /* CONFIG_DEBUG_PAGEALLOC */
  376. static int fixmaps;
  377. void __set_fixmap (enum fixed_addresses idx, phys_addr_t phys, pgprot_t flags)
  378. {
  379. unsigned long address = __fix_to_virt(idx);
  380. if (idx >= __end_of_fixed_addresses) {
  381. BUG();
  382. return;
  383. }
  384. map_page(address, phys, pgprot_val(flags));
  385. fixmaps++;
  386. }
  387. void __this_fixmap_does_not_exist(void)
  388. {
  389. WARN_ON(1);
  390. }